arqux 0.3.1

Enterprise Agent Infrastructure. Governance framework for AI agents with behavioral contracts, identity handoff, and automatic knowledge elevation.

⬡ ArqUX v0.3.1

Enterprise Agent Infrastructure.

Tags: ai-governance agent-orchestration llm-framework enterprise-ai mcp-server multi-agent audit-trail behavioral-contracts knowledge-management cortex-format

ArqUX is the first governance infrastructure purpose-built for AI agents in enterprise environments. It defines how an agent thinks, acts, learns, and accounts for its decisions — all under binding behavioral contracts that no command can override.

Built on a radical principle: the agent is the interface.

---

🔍 Use Cases

Scenario	How ArqUX Helps
AI agent in production	Traces every decision, enforces role separation, persists institutional memory
Multi-agent team	Each identity has its own behavioral contract (AXM, LIM, FCS); hot handoff between them
Compliance & audit	Every handoff, state transition, and mutation is recorded as verifiable evidence in brain.cortex
Knowledge retention	Lessons auto-elevate from individual → project → organization level without manual intervention
Agent governance	The framework governs itself (dogfooding) — every feature is a governed Blueprint
Enterprise deployment	71 MCP handlers, structured decision maps (18-section Blueprints), session resilience

---

⚡ Quick Start

pip install arqux
cd your-workspace
arqux init

The agent reads AGENTS.md, discovers context, and is operational. Everything else it learns on the fly.

---

🏛️ Architecture

Component	Purpose
Active Identity	Agent operating under a behavioral contract (AXM, LIM, FCS)
Handoff Manager	Switches identities via greeting or request
MCP Handlers	71 handlers for governance operations
brain.cortex	Project shared memory (decisions, lessons, state)
meta-brain	Consolidated workspace state across projects
Skills Library	Reusable workflows across agents and projects

---

🆔 Agent Identities — The Team

ArqUX replaces "one agent" with a team of specialized identities, each with its own binding behavioral contract.

Identity	Role	Scope
⬡ Alfred	Governor	Cycles, BLPs, tasks, approvals
⬡ Jarvis	Technical executor	Assigned tasks (claim, complete, evidence)
⬡ Seshat	Scribe	Documentation, diagrams, presentations
⬡ Heimdall	Guardian	Audit, monitoring, reporting

The Architect switches between identities with natural fluency:

"Hello Jarvis" → starts as technical executor "Switch to Seshat" → handoff to documentation "Back to Alfred" → resumes governance

Each identity has hard limits (LIM) that even the Architect cannot violate without an explicit handoff.

---

📋 Blueprints — The Decision Map

A Blueprint (BLP) captures every dimension of a design decision in 18 sections that the agent synthesizes autonomously after a single inquiry conversation.

§1 Problem → §2 Objective → §3 Preconditions → §4 Guiding Principle
§5 Context → §6 Scope → §7 Rules → §8 Technical Design
§9 Operational Design → §10 Contracts → §11 Work Procedure
§12 Acceptance Criteria → §13 Validations → §14 Tasks
§15 Risks → §16 Blocking Rule → §17 Expected Output → §18 Quality

Lifecycle

create → mature (inquiry + synthesis) → ready → claim
→ execution (checkpoint per task) → complete → verify → approve

---

🧠 Continuous Learning

Every lesson follows an automatic elevation pipeline:

Level	Description	Persists in
BEHAVIORAL	Individual agent lesson	identity.record() → agent .cortex file
CONTEXTUAL	Project-level knowledge	brain.cortex KNW
PROCEDURAL	Reusable workflow	Skill in .arqux/skills/

A lesson learned by one agent in one project can become a skill for every agent in the organization, without manual intervention.

---

🎯 Who ArqUX Is For

ArqUX is designed for organizations where:

• ✅ Compliance is not optional. Every action must be auditable.
• ✅ Agents operate continuously and must maintain coherence across sessions.
• ✅ Role separation is critical. Not every identity can do everything.
• ✅ Institutional knowledge must persist beyond ephemeral model memory.
• ✅ The cost of failure justifies investment in governance.

Not for afternoon experiments. For teams putting AI agents into production with real responsibilities.

---

🗺️ Roadmap

v0.3.x — Foundation (current)

• Governance framework with 71 MCP handlers and 22 BLPs in CYCLE-01
• CORTEX format with sigils, validation, and automatic elevation
• Multi-identity model with binding behavioral contracts
• Hot handoff between identities via greeting or request
• Autonomous Blueprint synthesis (inquiry → batch of 18 sections)
• Learning pipeline: BEHAVIORAL → CONTEXTUAL → PROCEDURAL
• Structural markers <!-- BLP:N --> for precise section replacement
• Full dogfooding — ArqUX governs itself

v1.1 — Operational maturity

• Native HCORTEX ↔ CORTEX translation
• Self-healing structural inconsistencies
• Consolidated workspace dashboard
• Quick-response tasks for lightweight actions

v2.0 — Enterprise scale

• Progressive mode: from basic operation to full governance
• Institutional skills marketplace
• CODEC-CORTEX as standalone library
• Interoperability with enterprise agent ecosystems
• Visual governance and audit interface

---

📚 Documentation

Resource	Description
AGENTS.md	Agent entry point — one file to start
protocol.skill.md	Session protocol, handoff, and decision-making
w08-blueprint-lifecycle.md	Complete Blueprint lifecycle
w10-identity-handoff.md	Hot identity handoff workflow
cortex.skill.md	CORTEX format and HCORTEX reference

---

⬡ ArqUX — The agent is the interface.